A method and system for identifying field roads

Abstract

This invention relates to a method and system for identifying field roads. The method includes: acquiring field image information using a drone; determining a segmentation threshold based on the field image information using the golden section method; segmenting the field image information using the segmentation threshold to obtain segmented roads; and determining the field roads using a mathematical cumulative model based on the segmented roads. This invention can quickly achieve automatic identification of farmland road information.

Description

Technical Field

[0001] This invention relates to the field of image recognition, and in particular to a method and system for recognizing field roads. Background Technology

[0002] The Importance of Basic Farmland Construction. The state attaches great importance to basic farmland construction and invests heavily in high-standard farmland development. Monitoring the quality of basic farmland construction is a key means of high-standard farmland construction, and effective monitoring of high-standard farmland construction in large-scale agriculture must be achieved through information technology.

[0003] Using drones for automated farmland information collection improves the efficiency of data acquisition and identification. Accessibility of farmland roads is a crucial standard for high-standard farmland; automatically calculating road accessibility necessitates the automated collection of road information. Drone remote sensing, with its advantages of all-weather operation, real-time monitoring, high resolution, flexibility, and cost-effectiveness, is playing an increasingly important role in various fields of national economic and social development, including agriculture, ecological environment, new rural construction planning, natural disaster monitoring, infrastructure security, water resources, and mineral resource exploration, becoming a new development direction following satellite remote sensing technology. Automated collection of farmland road information using drones will significantly improve the efficiency of farmland road information collection and identification.

[0004] Problems Existing in the Automatic Collection and Recognition of Farmland Road Information. Currently, the automatic extraction and recognition of road information in the transportation field are hot research topics. Many methods exist for road recognition, but all have various imperfections and unresolved shortcomings. Therefore, automatic road recognition has long been a focus of expert attention. The same problems exist for farmland roads, which have not yet received expert attention. Currently, the extraction and recognition of farmland road information is virtually nonexistent; therefore, the automatic acquisition and recognition of farmland road information is of even greater research significance.

[0005] The identification of field roads cannot be directly applied to existing urban road identification methods. The application needs and characteristics of field road information in high-standard farmland differ from those in urban areas. Therefore, the automatic identification of field roads cannot be directly applied to existing urban road identification methods. Currently, the application needs of urban road identification mainly include autonomous driving, navigation, intelligent transportation, and mapping, while the application needs of farmland roads are related to road accessibility. Urban road application systems can be equipped with various high-end devices, while the monitoring environment of farmland roads is poor and can only carry the most basic equipment. Urban roads can be set up in a better network environment, while farmland roads are mostly in a weak network environment. Urban road environments are relatively standardized, while farmland environments are harsh, and the characteristics of surrounding features differ. Urban traffic lines and farmland roads differ in color, shape, and texture. For monitoring high-standard farmland construction projects, and for large-scale farmland management and planning, due to the vast area of ​​farmland, the manpower, material resources, and time costs of traveling to and from the field for data collection are high. Therefore, farmland field data collection requires a system with rapid data collection and efficient on-site decision-making capabilities.

[0006] Therefore, the automatic extraction and identification of farmland road information is a new problem under the new situation of agricultural modernization.

[0007] Currently, there are many conventional methods for road identification, the most common being searching for relevant roads on Google Maps. Google Maps can provide road maps for various locations. However, for new agricultural infrastructure projects, Google Maps cannot immediately construct the roads for these new projects. Furthermore, satellite remote sensing imagery only passes over the area a limited number of times each year, and weather and cloud cover significantly impact image quality during each pass. Therefore, using satellite remote sensing to extract information on newly constructed roads for basic agricultural infrastructure in real time is impractical.

[0008] Remote sensing or drone imagery can only be viewed by the human eye; computers cannot directly identify roads. Computers can only recognize data, including data on fields, roads, greenery, and other features within the imagery. Computers cannot directly identify road data from the imagery. Therefore, it is necessary to filter out the complex non-road feature data to extract road data. Only after extracting the road information can computers analyze the road data and automatically assess the progress of road construction, such as road length and accessibility.

[0009] There is a wealth of research on extracting road information from road images, including traditional methods, machine learning, deep learning, and data fusion, all of which have achieved some success. However, most of these methods are only applicable to specific scenarios and needs. Furthermore, these methods still have many problems, such as image noise removal. Traditional noise removal methods mainly include mean filtering and median filtering. While these algorithms perform well in certain areas, they cannot effectively remove various noises superimposed on grayscale road images. Another example is morphological edge detection algorithms based on multi-scale structural elements, which are complex to implement, slow to process, and cannot meet the real-time requirements of road detection.

[0010] More importantly, many current methods remain at the theoretical research level. Transplanting these theoretical methods to application systems requires significant modifications and innovations to existing methods. While many road information extraction operators are relatively mature, their application conditions are highly specific, lacking universality and real-time performance. They also place high demands on network conditions and computer equipment, requiring auxiliary tools under specific computer configurations to complete calculations. For large-scale farmland construction projects with poor infrastructure, the overall design and methodology of road recognition systems must be thoroughly updated.

[0011] The application technology for field road identification is still very weak. To date, no successful software is available. The current feasible method is to manually collect images from the field and bring them back to the laboratory for manual identification using ArcGIS software. However, the identification results still have many shortcomings, and the process is time-consuming and labor-intensive. Therefore, existing field road identification technology is still far from being suitable for large-scale farmland project monitoring and on-site decision-making.

[0012] Road identification using existing Google remote sensing images still relies on manual visual inspection, and the image acquisition period is fixed, the distance is long, cloud cover is a problem, noise reduction is inadequate, and Google Maps cannot respond instantly to newly constructed roads. Existing road identification theories are predominantly theoretical research methods with limited application technologies. Current road identification applications suffer from poor real-time performance, high configuration requirements, poor practicality, and low levels of full automation.

[0013] In summary, most current road image recognition methods are geared towards high-end, large-scale, and sophisticated applications, and have not proposed convenient and effective road recognition methods for the specific needs of grassroots agriculture and the unique environment of farmland. Summary of the Invention

[0014] The purpose of this invention is to provide a method and system for identifying field roads, so as to quickly achieve automatic identification of farmland road information.

[0015] To achieve the above objectives, the present invention provides the following solution:

[0016] A method for identifying field roads includes:

[0017] Using drones to acquire field image information;

[0018] The segmentation threshold is determined using the golden section method based on the field image information.

[0019] The segmented roads are obtained by segmenting the field image information using the segmentation threshold.

[0020] Field roads are determined using a mathematical cumulative model based on the segmented roads.

[0021] Optionally, after acquiring field image information using a drone, the method further includes:

[0022] The field image information is processed to obtain a grayscale image;

[0023] Feature analysis is performed on the grayscale image to obtain the grayscale value range.

[0024] Optionally, determining the segmentation threshold using the golden section method based on the field image information specifically includes:

[0025] The segmentation threshold is determined using the golden section method based on the grayscale value range of the field image information.

[0026] Optionally, after segmenting the field image information using the segmentation threshold to obtain segmented roads, the method further includes:

[0027] The segmented roads are binarized.

[0028] Optionally, determining field roads using a mathematical cumulative model based on the segmented roads specifically includes:

[0029] A grayscale accumulation model is determined based on the continuous pixels of the segmented road; the continuous pixels include continuous pixels along the row direction and continuous pixels along the column direction.

[0030] Field roads are determined based on the grayscale value accumulation model.

[0031] A field road recognition system, comprising:

[0032] The acquisition module is used to acquire field image information using drones;

[0033] The segmentation threshold determination module is used to determine the segmentation threshold based on the field image information using the golden section method.

[0034] The segmentation module is used to segment the field image information using the segmentation threshold to obtain segmented roads;

[0035] The field road determination module is used to determine the field roads based on the segmented roads using a mathematical cumulative model.

[0036] Optionally, the field road identification system further includes:

[0037] The grayscale processing module is used to perform grayscale processing on the field image information to obtain a grayscale image.

[0038] The feature analysis module is used to perform feature analysis on the grayscale image to obtain the grayscale value range.

[0039] Optionally, the segmentation threshold determination module specifically includes:

[0040] The segmentation threshold determination unit is used to determine the segmentation threshold based on the grayscale value range of the field image information using the golden section method.

[0041] Optionally, the field road identification system further includes:

[0042] The binarization module is used to perform binarization processing on the segmented roads.

[0043] Optionally, the field road determination module specifically includes:

[0044] A grayscale accumulation model determination unit is used to determine a grayscale accumulation model based on the continuous pixels of the segmented road; the continuous pixels include continuous pixels along the row direction and continuous pixels along the column direction.

[0045] The field road determination unit is used to determine field roads based on the gray value accumulation model.

[0046] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0047] This invention utilizes drones to acquire field image information; determines a segmentation threshold based on the field image information using the golden section method; segments the field image information using the segmentation threshold to obtain segmented roads; and determines the field roads using a mathematical cumulative model based on the segmented roads. This invention uses drones to collect field images, performs road segmentation through thresholding, and uses a mathematical cumulative model to determine noise, thereby quickly achieving automatic identification of farmland road information. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 A flowchart of the field road identification method provided by this invention;

[0050] Figure 2 This is a schematic diagram of the field road identification method provided by the present invention;

[0051] Figure 3 (a) is the original map of the target site; Figure 3 (b) is a diagram showing the threshold segmentation results for the target plot; Figure 3 (c) The result image of the target plot after noise removal;

[0052] Figure 4 The image extraction results of the field road identification method and the ArcGIS method provided by this invention are shown in the figure. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] To address the current needs for monitoring large-scale standardized farmland construction projects, this invention utilizes drones to collect field information and analyzes the features of field features in the field images. A simplified method for identifying field roads is optimized and selected. A road segmentation threshold is chosen to perform binary image segmentation of roads. Then, a simplified denoising model is constructed to remove noise from the road images. The accuracy of the road identification results is verified using the kappa method. This invention achieves the goal of simple, easy, practical, and rapid extraction of farmland construction road information.

[0056] 1) Basic methods of road recognition:

[0057] Image segmentation refers to separating meaningful objects from their background in an image and dividing these objects according to their different meanings; in other words, it's about extracting objects with different meanings from the image. Image segmentation methods can be broadly classified into two categories: edge detection-based methods and region generation-based methods. Edge detection is the first step in all boundary-based image analysis methods. It first detects discontinuities in local image features and then connects them to form boundaries.

[0058] Edge detection methods address the discontinuities in local image characteristics (such as abrupt changes in grayscale, color, or texture) that mark the end of one region and the beginning of another. For computational convenience, first and second derivatives are typically used to detect boundaries, and differentiation methods can easily detect discontinuities in grayscale values. Classical algorithms such as gradient operators, Robert gradients, and Laplacian operators can be employed.

[0059] Region segmentation is the most basic method for separating meaningful regions in a grayscale image. It involves setting a threshold. Examples include the bimodal method, the p-parameter method, and the maximum variance automatic thresholding method.

[0060] 2) Several commonly used techniques for road recognition: Binary images and mathematical morphology.

[0061] Binary images: For a long time, people have been exploring how to directly obtain the regions or edges that constitute the shape of an object from a grayscale image, resulting in various effective methods. Especially for the recognition and understanding of complex scene images, the rich grayscale information in the image becomes the main basis; an image with only two grayscale levels is often sufficient to study the shape characteristics of an object. Compared with grayscale images, binary images contain significantly less information, thus processing an image is faster, less costly, and has high practical value.

[0062] Mathematical morphology: Shape feature technology is the process of further abstracting shape feature parameters based on image segmentation, binarization, and line drawing. Morphology, or mathematical morphology, is one of the most widely used techniques in image processing. It is mainly used to extract image components that are meaningful in expressing and describing the shape of regions, enabling subsequent recognition work to capture the most essential (most discriminative) shape features of the target object, such as boundaries and connected regions. There is no unified definition for shape feature parameters; any parameter that can fully reflect the shape of an object, effectively distinguish the shape differences between objects, and is easily and quickly obtained can be used as a shape feature parameter. Therefore, in practical applications, there are many ways to describe it; regions and edges can be described directly, and even series of feature parameters can be generated through mathematical methods.

[0063] like Figure 1 As shown, the present invention provides a method for identifying field roads, comprising:

[0064] Step 101: Use drones to acquire field image information.

[0065] After acquiring field image information using a drone, the method further includes: performing grayscale processing on the field image information to obtain a grayscale image; and performing feature analysis on the grayscale image to obtain a grayscale value range.

[0066] Step 102: Determine the segmentation threshold using the golden section method based on the field image information. Step 102 specifically includes: determining the segmentation threshold using the golden section method based on the grayscale value range of the field image information.

[0067] Step 103: Segment the field image information using the segmentation threshold to obtain segmented roads.

[0068] After segmenting the field image information using the segmentation threshold to obtain segmented roads, the method further includes: performing binarization processing on the segmented roads.

[0069] Step 104: Determine field roads using a mathematical cumulative model based on the segmented roads.

[0070] Step 104 specifically includes:

[0071] A grayscale accumulation model is determined based on the continuous pixels of the segmented road; the continuous pixels include continuous pixels along the row direction and continuous pixels along the column direction.

[0072] Field roads are determined based on the grayscale value accumulation model.

[0073] When consecutive pixel values ​​in a grayscale accumulation model are all the same, it indicates that the pixel is a road pixel.

[0074] The present invention also provides a field road recognition system, comprising:

[0075] The acquisition module is used to acquire field image information using drones.

[0076] The segmentation threshold determination module is used to determine the segmentation threshold based on the field image information using the golden section method.

[0077] The segmentation module is used to segment the field image information using the segmentation threshold to obtain segmented roads.

[0078] The field road determination module is used to determine the field roads based on the segmented roads using a mathematical cumulative model.

[0079] In practical applications, field road recognition systems also include:

[0080] The grayscale processing module is used to perform grayscale processing on the field image information to obtain a grayscale image.

[0081] The feature analysis module is used to perform feature analysis on the grayscale image to obtain the grayscale value range.

[0082] In practical applications, the segmentation threshold determination module specifically includes: a segmentation threshold determination unit, used to determine the segmentation threshold based on the grayscale value range of the field image information using the golden section method.

[0083] In practical applications, the field road recognition system also includes a binarization processing module, which is used to perform binarization processing on the segmented roads.

[0084] In practical applications, the field road determination module specifically includes:

[0085] The grayscale accumulation model determination unit is used to determine the grayscale accumulation model based on the continuous pixels of the segmented road; the continuous pixels include continuous pixels along the row direction and continuous pixels along the column direction.

[0086] The field road determination unit is used to determine field roads based on the gray value accumulation model.

[0087] This invention prioritizes applicability, adhering to the principles of saving time, manpower, and configuration. It adopts a simplicity-first design philosophy, starting with region segmentation and using the simplest golden section method to design the optimal threshold. This directly binarizes the grayscale image, significantly reducing the amount of information, resulting in faster image processing, lower cost, and greater practical value. It abandons existing denoising operators and constructs a simple mathematical model of linear features, greatly saving computation time. Figure 2 As shown, the specific steps are as follows:

[0088] (1) Optimization and selection of field information collection, analysis and identification methods, including UAV field road information collection, field road information feature analysis and optimization and selection of simplified field road identification methods.

[0089] Unmanned aerial vehicle (UAV) field information collection

[0090] Time period - multiple time periods from December 31, 2020 to April 23, 2021; Location - Ningxi Base of South China Agricultural University, Zengcheng, Guangzhou, Guangdong Province; Weather - sunny; Drone - DJI Phantom P4 Multispectral; Camera model - P4Mcamera; Camera resolution - (3.2--9.3cm / px); Flight altitude - (73.3m---214.1m); Coverage area - (5.25 acres--41.1 acres).

[0091] like Figure 3 As shown in (a), an image of the target plot is selected, and the pixel grayscale values ​​of the image are extracted using Phython general software. In this invention, only the grayscale values ​​of the B image in the RGB image are used for analysis and identification. The results of the field information feature analysis are shown in Table 1.

[0092] Table 1. Pixel Features of Field Features

[0093]

[0094]

[0095] Analyze the imagery acquired by the drone (including R, G, and B imagery), and extract road information from image B. For example... Figure 3 (a) and Figure 3 As shown in (b), it can be seen from the gray values ​​of the image pixels that the gray values ​​of bare soil, plastic film and road are similar, making it difficult to extract road information from them using thresholding. The gray values ​​of other land features are significantly different from the gray values ​​of roads, and road can be segmented from them using thresholding.

[0096] Therefore, the key to segmenting farmland roads using thresholding is to accurately define the grayscale boundaries of shadows, bare soil, and plastic film. This is why the case study specifically selected fields containing these typical features (see...). Figure 3The road is flanked by two plots of farmland with bare or semi-bare soil. The roadside is covered with weeds and bare soil. There is a plastic film covering the top left corner of the picture.

[0097] (2) A simplified method for automatic identification of field roads, including road threshold selection, road binarization image segmentation, construction of a simplified mathematical morphology denoising model, and road image noise processing.

[0098] With simplicity as the goal, an online monitoring system for high-standard farmland construction projects is constructed that is fast, efficient, and requires minimal configuration. Method optimization selection:

[0099] Method 1:

[0100] The goal of this invention is automatic road recognition. Therefore, edge detection operators such as Robert gradient or Laplacian can be used to automatically identify road edges, segment the road image, and then use mathematical morphology to denoise the image, thus extracting road information. This requires calling a high-level visual recognition operator library to automatically identify road edges using edge detection operators such as Robert gradient or Laplacian. This will reduce the speed of the recognition process and also increase the level of equipment configuration and the difficulty of installing application software.

[0101] Method 2:

[0102] Given the system's purpose, and to achieve the goal of fast, efficient, and low-configuration online monitoring of large-scale field projects, this invention adopts a 'simplicity-first' strategy. First, the grayscale value characteristics of typical farmland features are analyzed, and a grayscale value feature-first strategy is selected. A threshold is set to directly segment the pixels of the target road information. Then, morphological features are used to address the fragmented and discontinuous morphological characteristics of farmland road noise, establishing a simplified connectivity recognition model to remove noise and extract road information.

[0103] Since the design philosophy prioritizes simplicity, this invention will adopt the simplified approach of Method Two. First, a grayscale-threshold approach is used to target the road, immediately filtering out most non-target pixels such as crops and vegetation. Then, the high-level computations of existing classic edge detection operators like Laplacian are abandoned, and a simplified mathematical denoising model based on accumulation is constructed. This achieves the goal of rapid and efficient monitoring using low-configuration equipment while maintaining a simplified model. The road recognition process is as follows:

[0104] I. Threshold segmentation of roads --- Threshold segmentation of pixel grayscale values.

[0105] Based on the color characteristics of farmland roads and surrounding features, the optimal threshold for segmenting gray-white pixels on roads was determined. Typical road surfaces in standard farmland are made of cement, asphalt, or a mixture of cement, asphalt, and mortar. Farm roads are 2.5-3 meters wide, while production roads are 2-2.5 meters wide. The production base in this invention uses cement-paved roads. The grayscale values ​​of pixels (including grayscale values ​​from RGB images) can be obtained from farmland images using Python. This invention only uses the grayscale values ​​of the B (blue) image (RGB).

[0106] As can be seen from the field information feature analysis table, the largest proportion of the acquired field pixels is the crop vegetation cover. The gray value of the crop vegetation cover is significantly different from that of the road surface. Therefore, they are easy to segment using a threshold. After setting the threshold, most of the non-target pixels were removed, leaving only a small number of pixels (bare soil, plastic film, and shadows) with gray values ​​similar to those of the road surface.

[0107] This invention can identify bare soil, greenhouses, shaded areas, and road surfaces with grayscale values ​​between 80 and 200. It employs the simplest golden ratio method (0.618 segmentation) to quickly search for the optimal threshold shown in Table 2 within the 80-200 grayscale value range. The invention selects... Figure 3 (a) shows a typical image containing large areas of various types of bare soil, crop vegetation, plastic film and road surface.

[0108] Table 2. Optimal Threshold Calculation Using the Golden Section Method

[0109]

[0110] Comparing the segmentation results of several rounds of experiments, the optimal threshold was found to be 139. After binarization (if the pixel grayscale value in image B is >139, then the pixel is set to white with an RGB grayscale value of -255, 255, 255; otherwise, the pixel is set to black with an RGB grayscale value of -0, 0, 0), from... Figure 3 As can be seen in (b), the target road of the project was extracted (white), and the green vegetation on both sides of the road was filtered out (black). However, a lot of noise remains in the image. The bare soil and semi-bare soil in the farmland, as well as the plastic film covering the fields, were extracted to varying degrees, forming many white spots.

[0111] The threshold optimized by this invention is suitable for general conditions such as sunny days (or few clouds) and cement roads in the fields. If other road materials are used, on-site shooting can be conducted before aerial photography, and the optimal threshold can be determined using the golden ratio method.

[0112] II. Mathematical Noise Reduction Model – Mathematical Expression of Continuous Features of Road Pixels

[0113] Analysis of road and noise morphological characteristics. Figure 3(b) It can be seen from the sub-figure that noise is mostly in the form of fine, salt-and-pepper-like or thin strips, while roads are continuous and wide strips, which is the difference in their morphological characteristics.

[0114] Mathematical representation of the 'discontinuous' characteristics of noise. After thresholding and binarizing the road, the grayscale values ​​of road pixels are (255, 255, 255 - white) (represented by white in the attached diagram), while the grayscale values ​​of other non-road pixels are (0, 0, 0 - black). Therefore, the sum of the grayscale values ​​of any non-road pixel and its adjacent pixels will always be less than the sum of the road pixel values. This is because the road pixel grayscale values ​​are a continuous series of 255s, while the non-road pixel grayscale values ​​are mostly 0s or discontinuous 255s. Therefore, by summing the grayscale values ​​of each pixel and its adjacent pixels, the sum of the road pixel values ​​is approximately 255*N, where N is the number of pixels accumulated. The sum of the non-road pixel values ​​is much less than 255*N, effectively filtering out salt-and-pepper or stripe-like noise. Performing column-wise accumulation first, followed by directional accumulation, can eliminate salt-and-pepper or stripe-like noise in any direction.

[0115] Mathematical noise cancellation models:

[0116] grayscale value x pj The pixel grayscale value accumulation model along the row direction:

[0117] grayscale value x ip The pixel grayscale value is accumulated along the column direction using the following model:

[0118] Among them, S pj For the grayscale value accumulation model along the row direction, x pj S represents the grayscale value. ip For the grayscale value accumulation model along the column direction, x ip This is the grayscale value.

[0119] Table 3 is a table of binarized data after the road is segmented using thresholds, such as... Figure 3 As shown, the large areas of 0 on the left represent the separated bare soil fields, while the rows of 255 on the right represent roads. Because of the varied colors within the bare soil fields, some loose soil or shadows cannot be segmented using the thresholding method, resulting in salt-and-pepper or thin stripe-like noise, such as pixels -119 and P in Table 3. The following section will use a morphological accumulation model to distinguish between noise and roads.

[0120] Table 3. Binarized data after road segmentation.

[0121]

[0122] Take the pixel (119, AC) in the road, then

[0123] grayscale value x 119,AC Pixel grayscale value accumulation model along the row direction

[0124]

[0125] grayscale value x 119,AC Pixel grayscale value accumulation model along column direction

[0126]

[0127] Take the non-road pixels (120, P)

[0128] grayscale value x 120,P Pixel grayscale value accumulation model along the row direction

[0129]

[0130] grayscale value x 120,P Pixel grayscale value accumulation model along column direction

[0131]

[0132] It can be seen that pixel (119, AC) is a road pixel because the row and column connected to it are all consecutive pixels of 255. Therefore, the grayscale value of this pixel can be set to (255, 255, 255 - white), and the individual pixels of the road are extracted. It can be seen that point (120, P) is noise because the row (or column) connected to it is not consecutive pixels of 255. Therefore, the value of this pixel can be set to (0, 0, 0 - black), and the individual noise pixels are eliminated.

[0133] Therefore, if the cumulative value of any pixel's row and column matches the shape of a road (pixels in both rows and columns are continuous), then it is a pixel representing a road; otherwise, it is noise. For example... Figure 3 As shown in (c), it can be seen that most of the road pixels have been extracted. Although noise elimination is sometimes not ideal, other methods also have various defects. The method provided by this invention is efficient, fast and can basically meet the subsequent calculation requirements of road length and accessibility.

[0134] The noise cancellation model of this invention is suitable for drones, cameras, and weather conditions under general conditions. The drone used in this invention (DJI P4 Multispectral), camera (P4M camera), resolution (3.2-9.3 cm / px), and clear or partly cloudy weather conditions, with a flight altitude of 73.3 meters to 214.1 meters, require that the number of accumulations m and n in the model be greater than the number of pixels with the largest noise pattern. If conditions change, the number of accumulations m and n can be adjusted to achieve the purpose of noise cancellation.

[0135] (3) Results and Analysis, Kappa Road Recognition Accuracy Verification

[0136] This invention studies images of Zengcheng base taken by drones at different altitudes (73.9 meters to 214.1 meters), at different time periods, and in different batches, demonstrating that the method of this invention has a certain degree of universality.

[0137] The farmland road images extracted using the method of this invention are then used to extract farmland road data information using a simplified road recognition model. Figure 4 This is a partial result of road information extraction. Typical farmland features include: most of the area is covered by green crops, with smaller areas consisting of roads, bare soil, plastic film, shade netting, and irrigation ditches. Bare soil and plastic film are similar in color to roads.

[0138] Figure 4 (a) is the original image of the first image. Figure 4 (b) is the extraction result of the first original image using a simplified method. Figure 4 (c) is the result of extracting the first original image using the ArcGIS method. Figure 4 (d) is the original image of the second image. Figure 4 (e) is the extraction result of the second original image using a simplified method. Figure 4 (f) shows the extraction result of the second original image using the ArcGIS method. Figure 4 (g) is the original image of the third image. Figure 4 (h) is the extraction result of the third original image using the simplified method. Figure 4 (i) is the result of extracting the third original image using the ArcGIS method. Figure 4 (j) is the original image of the fourth image. Figure 4 (k) is the extraction result of the fourth original image using the simplified method. Figure 4 (l) is the result of extracting the fourth image using the ArcGIS method. Figure 4 (m) is the original image of the fifth image. Figure 4 (n) represents the extraction result of the fifth original image using a simplified method. Figure 4(o) is the extraction result of the fifth original image using the ArcGIS method, where the simplified method is the recognition method provided by this invention. From Figure 4 (a)- Figure 4 The image (o) shows that the road is well separated from the green cover and shade net, but the separation from the bare soil and semi-bare soil is not ideal in some parts.

[0139] Compare the classification results of the two methods with the original image:

[0140] From the overall classification accuracy of the five groups using the two classification methods in Table 4, most of them have a high overall classification accuracy (above 90%), indicating that the overall classification results of the two methods are good. Only... Figure 4 (f) The overall classification accuracy of the ArcGIS method is only 85%. (See attached image.) Figure 4 ArcGIS subgraphs Figure 4 (f) shows that its denoising results are particularly poor. Therefore, the overall classification accuracy value is consistent with the actual situation.

[0141] Table 4. Kappa consistency test results of the two road information extraction methods compared with the original map.

[0142]

[0143]

[0144] From the kappa coefficients of the two classification methods in the five groups in Table 4, Figure 4 (a) and Figure 4 (d) all have low kappa coefficients. (See attached data.) Figure 4 From the corresponding sub-images, it can be seen that their denoising effects are all poor, especially... Figure 4 (d) The ArcGIS classification results for the sub-image are even worse; large areas of bare soil along the roadside and many field ridges cannot be denoised. Therefore, the kappa coefficient is consistent with the actual situation.

[0145] According to the Kappa consistency test, the simplified method of the present invention has higher consistency with the original image. The simplified method extracts farmland road information to the same level as the ArcGIS method, and sometimes even exceeds its results.

[0146] As shown in Table 5, the overall comparison of the two road information extraction methods shows that the simplified method has an absolute advantage in meeting the requirements of rapid and automated detection of farmland roads on a large scale. Its biggest drawback is that the visual effect is not ideal, but the rapid and automated detection process of farmland roads does not require human visual intervention.

[0147] Table 5. Overall Comparison of Two Road Information Extraction Methods

[0148]

[0149] This invention aims for simplicity, speed, efficiency, and practicality. While its visual quality is lower than ArcGIS software, its road information acquisition and recognition results are no less than those achieved with ArcGIS. Its equipment requirements are significantly lower than ArcGIS's, its operation is less complex, and its operation time is more than 1 / 14th that of ArcGIS. It boasts a higher degree of automation than ArcGIS, achieves or exceeds ArcGIS's accuracy, has an instantaneous response time of 0.25-0.5 minutes, and offers far greater potential for secondary development. The proposed method utilizes the characteristics of low-altitude remote sensing imagery, employs threshold segmentation to extract road information, and then uses morphological methods for noise reduction and post-processing to obtain the final extracted results. This method can be effectively applied to road extraction from low-altitude remote sensing imagery. However, the visual quality is not ideal, and the results are primarily for computer recognition, not manual recognition. Furthermore, the recognition of areas with shadows or large greenhouses is similar to ArcGIS, failing to achieve ideal results. Further research and verification are needed regarding its generalizability.

[0150] This invention is applicable under the following conditions: flight altitude 73.9-214 meters, resolution 3.2-9.3 cm / px, clear skies with few clouds. If weather conditions are unsatisfactory or road color changes, the calibration can be repeated. The threshold can be adjusted and tested to improve road extraction performance. Therefore, this invention uses the simplest golden ratio method to calibrate the threshold. If the image resolution changes, the calibration can be repeated. The number of accumulations can be adjusted and tested to ensure even the smallest roads are displayed.

[0151] This invention utilizes drones to collect field information and analyze field feature characteristics in the images, optimizing and selecting a simplified method for field road identification. A road segmentation threshold is selected for binarized image segmentation of field roads, followed by the construction of a simplified denoising model to remove noise from the road images. This achieves the goals of simplicity, ease of use, practicality, and rapid extraction of road information. The invention prioritizes practicality, simplicity, and speed. It employs the most basic equipment and simplest operating procedures, using fundamental image recognition methods to construct a drone-based system for acquiring and identifying high-standard farmland road information online.

[0152] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0153] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying field roads, characterized in that, include: Using drones to acquire field image information; The segmentation threshold is determined using the golden section method based on the field image information. The segmented roads are obtained by segmenting the field image information using the segmentation threshold. The field roads are determined using a mathematical cumulative model based on the segmented roads, specifically including: determining a grayscale value cumulative model based on the continuous pixels of the segmented roads; the continuous pixels include continuous pixels along the row direction and continuous pixels along the column direction; and determining the field roads based on the grayscale value cumulative model.

2. The field road identification method according to claim 1, characterized in that, Following the acquisition of field image information using drones, the following is also included: The field image information is processed to obtain a grayscale image; Feature analysis is performed on the grayscale image to obtain the grayscale value range.

3. The field road identification method according to claim 2, characterized in that, The step of determining the segmentation threshold using the golden section method based on the field image information specifically includes: The segmentation threshold is determined using the golden section method based on the grayscale value range of the field image information.

4. The field road identification method according to claim 1, characterized in that, After segmenting the field image information using the segmentation threshold to obtain segmented roads, the method further includes: The segmented roads are binarized.

5. A field road identification system, characterized in that, include: The acquisition module is used to acquire field image information using drones; The segmentation threshold determination module is used to determine the segmentation threshold based on the field image information using the golden section method. The segmentation module is used to segment the field image information using the segmentation threshold to obtain segmented roads; The field road determination module is used to determine the field roads based on the segmented roads using a mathematical cumulative model; The field road determination module specifically includes: A grayscale accumulation model determination unit is used to determine a grayscale accumulation model based on the continuous pixels of the segmented road; the continuous pixels include continuous pixels along the row direction and continuous pixels along the column direction. The field road determination unit is used to determine field roads based on the gray value accumulation model.

6. The field road identification system according to claim 5, characterized in that, Also includes: The grayscale processing module is used to perform grayscale processing on the field image information to obtain a grayscale image. The feature analysis module is used to perform feature analysis on the grayscale image to obtain the grayscale value range.

7. The field road identification system according to claim 6, characterized in that, The segmentation threshold determination module specifically includes: The segmentation threshold determination unit is used to determine the segmentation threshold based on the grayscale value range of the field image information using the golden section method.

8. The field road identification system according to claim 5, characterized in that, Also includes: The binarization module is used to perform binarization processing on the segmented roads.

Images